Journal of Cleaner Production - ae.ruc.edu.cn

lable at ScienceDirect

Journal of Cleaner Production 222 (2019) 573e583

Contents lists avai

Journal of Cleaner Production

journal homepage: www.elsevier .com/locate/ jc lepro

Efficiency evaluation of thermal power plants in China based on theweighted Russell directional distance method

Jun Xie a, Zhuang Liang a, Xiaobing Zhang b, Lei Zhu a, *

a School of Economics and Management, Beihang University, Beijing, 100191, Chinab School of Economics, Renmin University of China, Beijing, 100872, China

a r t i c l e i n f o

Article history:Received 3 May 2018Received in revised form12 January 2019Accepted 7 March 2019Available online 9 March 2019

Keywords:Thermal power plantsOverall efficiencyWeighted Russell directional distancemethodData envelope analysis (DEA)

* Corresponding author.E-mail address: [email protected] (L. Zhu

https://doi.org/10.1016/j.jclepro.2019.03.0780959-6526/© 2019 Elsevier Ltd. All rights reserved.

a b s t r a c t

China has the largest thermal power installed capacity worldwide, which has caused severe greenhousegas emissions. To reduce the emissions and facilitate the development of energy savings and emissionabatement, it is imperative to improve power-generating efficiency. In this paper, the nonparametricweighted Russell directional distance method (WRDDM) was used to evaluate the overall efficiency ofthermal power plants in 30 provinces in China. The results show that the efficiency of East and NorthChina is higher than that of Northeast and Southwest China. Furthermore, the correlations between scale,ownership and efficiency are discussed. The nonstate-owned plants with large generation capacity havea higher efficiency than the small and state-owned plants.

© 2019 Elsevier Ltd. All rights reserved.

1. Introduction

Energy savings and emission abatement of the power sectorplay essential roles in sustainable development. China has thelargest installed capacity of thermal power in the world, and themassive use of thermal power has led to excessive greenhouse gasemissions (IEA, 2018).

In addition, China's national Emission Trading Scheme (ETS) wasofficially established at the end of 2017, and the electric powerindustry was the first industry to be included. Based on this,improving the overall efficiency of the power sector is the bestalternative for achieving emission abatement before any techno-logical breakthroughs become operational (e.g., carbon capture andstorage, CCS). Therefore, the efficiency evaluation of thermal powerplants becomes significant, and such analysis can present in-structions on efficiency improvement for stakeholders andpolicymakers.

In this paper, a nonparametric approach of data envelopeanalysis (DEA), called the weighted Russell directional distancefunction model (WRDDM), was adopted to evaluate the overallefficiency of thermal power plants in 30 provinces in China from

).

2012 to 2014. Nonparametric methods refer to those that do notrequire a specific equation related to the inputs and outputs (Weiet al., 2013), and a key advantage of the WRDDM over other DEAmodels is that it can increase the expected outputs and decreasethe inputs as well as the unexpected outputs simultaneously andnonproportionally. Additionally, it can estimate the efficiency ofeach input and output and obtain the weighted overall efficiency.By using this method, decisionmakers can know specifically whichpart has been fully utilized or controlled and which part has po-tential for further improvement. Essentially, the overall efficiencyof each plant varies due to differences in the region, technology, themanagement level, the year of operation, etc. By comparing theoverall efficiency of different plants among different years and re-gions, the main factors that cause efficiency distinction can bedetermined. With the nonparametric test, we explored the de-terminants that may influence efficiency, such as the scale of plantsand the difference of ownership. The results can generate morecomprehensive suggestions on domestic plant efficiencyimprovement.

This article is organized as follows: Section 2 presents a litera-ture review of previous studies. Section 3 introduces the method-ology and data. In section 4, we illustrate the empirical results.Finally, we show the conclusions of our work and draw some policyimplications.

mailto:[email protected]

http://crossmark.crossref.org/dialog/?doi=10.1016/j.jclepro.2019.03.078&domain=pdf

www.sciencedirect.com/science/journal/09596526

http://www.elsevier.com/locate/jclepro

https://doi.org/10.1016/j.jclepro.2019.03.078



J. Xie et al. / Journal of Cleaner Production 222 (2019) 573e583574

2. Literature review

DEA is a nonparametric method for efficiency evaluation. Thismethod can estimate the effective production frontier based on aset of observations and obtain the production inefficiency of eachdecision-making unit. Zhou et al. (2008) collated nearly 100representative papers on DEA used in energy and environmentfields. Zhang and Choi (2014) summarized the directional distancefunction method from 1997 to 2013. F€are et al. (2007) used DEA tostudy the energy efficiency of American power plants. Zhang andChoi (2013) used DEA to evaluate the production and emission ef-ficiency of Korean fossil fuel power plants. Sueyoshi et al. (2011)used the nonradial DEA method to evaluate the energy efficiencyof Japan's fossil fuel power plants from 2005 to 2008 and comparedtheir environmental benefits with those of the manufacturing in-dustry in 2012 (Sueyoshi and Goto, 2011).

In studies on the power generation efficiency of China, Teng(2003) obtained the technical efficiency based on data fromChina's coal-fired power plants in 1991. Li and Zou (2012) selectedthe panel data of the power sector in 27 provinces from 2000 to2009 and evaluated the technical efficiency and further discussedthe influence of the economic level, the utility rate of electricity andincentive regulation on power generation efficiency. Choi et al.(2012) used the DEA model to estimate the CO2 emission abate-ment cost, the potential and the energy utilization efficiency ofthermal power plants based on data from 30 provinces from 2001to 2010. Zhang and Xia (2011) combined DEA with the stochasticfrontier approach (SFA) to analyze the technical efficiency of thepower generation industry based on the panel data of 30 provincesfrom 2003 to 2009. They found that the technical efficiency wasconsiderably low, and the state-owned plants performed worsethan the nonstate-owned plants. Xie et al. (2012) used a two-stageDEA model to study the relation between generation methods andthe environmental performance of the power industry and foundthat the correlationwas significant. Yang andMichael (2009) used athree-multiplication DEA model to evaluate the environmentalbenefits of 582 thermal power plants in China in 2002. Niu (2017)used a three-stage super-DEA model that took unexpected outputsinto consideration to calculate the carbon emission and intensityper capita with the data of fossil fuel consumption from 2003 to2013.

Apart from the power industry, DEA has also been used in otherenergy-intensive industries. Jerry et al. (2012) estimated the effi-ciency of the Swiss paper industry in 1995, 2000 and 2005, theyfound that technological progress and an increase in funds couldimprove efficiency. Mandal (2010) used DEA to explore the impactof an unexpected output and environmental policies on the energyutilization efficiency of the India cement industry from 2001 to2004, and he proposed that it would lead to great deviationwithoutconcerning the unexpected output. Liu and Wang (2015) improvedthe two-stage network DEA model and obtained an efficiencydecomposition DEA model to evaluate the industrial sector effi-ciency of all the provinces in China in 2008 and compared the re-sults with those of traditional DEA and the original two-stagenetwork DEA model.

In this paper, the WRDDM was applied to obtain the overallefficiency of each power plant. The difference between theWRDDMand other DEA models is that it takes unexpected outputs intoconsideration and can measure the efficiency of each input andoutput. Using WRDDM, we can not only evaluate the efficiency butalso show plants specifically how to improve their efficiency, forexample, by decreasing the number of employees or investing lesson equipment. This approach is more comprehensive than thetraditional DEA model, which is why WRDDM is more useful foroverall efficiency evaluation.

To the best of our knowledge, only a few scholars have appliedthe WRDDM to evaluate efficiency at the microlevel. Chen et al.(2011) first proposed this method on the basis of Chung et al.(1997) and Fukuyama and Weber (2009); then, Chen et al. (2015)improved this model using the data of 99 countries from 1991 to2003 and evaluated the efficiency of each input and output. Weiet al. (2015) used this method to calculate the emission abate-ment potential of 68 and 124 power plants in 2004 and 2008 inZhejiang Province, respectively, and constructed an energy-savingindex and emission-abatement index. María et al. (2016) usedthismethod to evaluate the environmental efficiency of wastewatertreatment plants. Our work is the first to apply this method withrecent data (2012 and after) in the power-generating industry inChina. Therefore, the results of this paper could be a usefulreference.

3. Data and methodology

3.1. Construction of the output set

The outputs of electricity generation are often divided into ex-pected outputs and unexpected outputs (Chung et al. (1997), Zhouet al. (2012), Wei et al., (2015), Yang and Michael (2010). In thispaper, vector x¼ {x1,x2,x3 x2Rþg was used to denote three inputs:labor, assets and energy consumption, y2Rþ denotes expectedoutputs and b2Rþ denotes unexpected outputs. P(x) representsthe possible output set, and P(x)¼ {(y, b), x can produce y, b}.

Meanwhile, P(x) satisfies the following assumptions (Wei et al.,2015).

(1) P(x) is a standard convex and compact set. That is, limitedinputs can only produce limited outputs.

(2) The input is freely disposable, which indicates that theoutput set will not contract when the input increases. That is:If x’>x, then P(x’)JP(x).

(3) The null-jointness axiom is satisfied, which implies thatunexpected outputs are jointly produced with expectedoutputs. That is: If (y, b)2P(x) and b¼ 0, y¼ 0.

(4) The expected and unexpected outputs satisfy joint weakdisposability, which means two kinds of outputs can bereduced proportionally by q. That is: If (y,b)2P(x) and0 � q � 1, then (qy,qb)2P(x).

(5) The expected output is strongly disposed, which implies thatit is possible to dispose the expected products withoutreducing the unexpected outputs. That is: If (y,b)2P(x), and(y’, b)�(y,b), then (y’, b)2P(x).

3.2. Directional distance function

The traditional directional distance function (DDF) can bedefined as follows:

D0�!�

x; y; b; gx; gy; gb�¼ sup

nb

:�yþ bgy;bþ bgb

�2Pðxþ bgxÞ

o

g is a direction vector, and g¼(-gx, gy, -gb) (g2RNþ � RMþ � RJþ). Thisindicates that it is possible to both increase expected outputs andreduce unexpected outputs and inputs. The direction for input andthe expected and unexpected outputs are gx, gy and gb, respectively.Based on the direction vector g and output set P(x), we can derivethe distance between the observations and the frontier, or the in-efficiency level b. When b¼ 0, the output of this observation lies on

J. Xie et al. / Journal of Cleaner Production 222 (2019) 573e583 575

the frontier, and the efficiency has reached the best outcomecompared with all the other observations. The higher b is, the lowerthe overall efficiency and the higher the potential for efficiencyimprovement.

We applied the WRDDM to evaluate the inefficiency of eachspecific factor and obtained the weighted overall inefficiency. Somemodifications were made based on the traditional DDF, and thenew model is as follows:

D0�!ðx; y;b; gÞ ¼ max

�uk0b

k01

1 þ uk0bk02

2 þ uk0bk03

3 þ uk0bk04

4 þ uk0bk05

5

�

s.t.

XKk¼1

zkyk ��yk

0 þ bk0

4 ,gy�

(1)

XKk¼1

zkbk ¼�bk

0 � bk0

5 ,gb�

(2)

XKk¼1

zkxkn ��xk

0

n � bk0

n ,gxn

�; n ¼ 1;2;3 (3)

XKk¼1

zk ¼ 1 (4)

zk � 0;k ¼ 1;2;&;K (5)

where zk denotes the weight given to each input and output toconstruct the possible output set. The two sides for the constraints,(i), (ii) and (iii), indicate the optimally performed plant and theactual observation. The normalized weight assigned for each inputand output is u¼(u1,u2,u3,u4,u5). The inefficiency for the threeinputs, the good output and the bad output, respectively, aredenoted by bnðn ¼ 1;2;3;4;5Þ. An inequality sign in (i) and (iii)indicate strong disposability in assumptions (2) and (5). Theconstraint (ii) presents the weak disposability of the unexpectedoutputs and the assumption of null-jointness. Constraint (iv) rep-resents that the model is under the assumption of various returnsto scale (VRS). In recent years, some studies have been conductedunder a constant return to scale (Wang et al., 2016; Han et al., 2018)and have been comparatively appropriate in terms of only gener-ating efficiency. In this paper, we aimed to obtain the efficiency foreach specific input and output, and it is reasonable to assume thatthe outputs would be enhanced by more than the proportionalchange in the inputs, such as labor, assets and energy consumption(Picazo. et al., 2005). Constraint (v) is used to ensure that all theintensity variables are nonnegative.

In this paper, CO2 emissions were the only unexpected outputincluded because greenhouse gas emissions are the primaryconcern. Meanwhile, we assumed that there were no priorities forthe different inputs or outputs, and thus, they had the sameweight.Following the previous study, u can be set as (1/9,1/9,1/9,1/3,1/3)(Zhang and Choi, 2013) and g¼(gx, gy, gb)¼(-xn, y, -b) (Wei et al.,2015), which means that the plants can scale the input/output inthe direction of g.

3.3. Data

In this paper, the annual average number of employees, the totalyear-end assets, the total energy consumption (coal equivalence),

the generating capacity and the CO2 emissions were applied to themodel. All the datawere acquired from the China Electricity Council(CEC). Due to data limitation, CO2 emissions were calculated bymultiplying the energy consumption by the emission factorreleased by IPCC in 2006. Considering the technique differences, weseparated the samples into two groups: coal-fired power plants andgas-fired power plants. Table 1 and Table 2 show the relevantvariables used in this model.

Fig. 1 shows the number of coal-fired plants in each province.The number of gas-fired plants is comparatively small (30, 35 and31). Provinces such as Hebei, Shanxi, Inner Mongolia, Jiangsu,Shandong, Henan, and Guangdong have a considerable number ofplants, but the samples in Jiangxi, Hainan, and Qinghai are notsufficient, which makes the efficiency evaluated less supportive.

4. Empirical results

The linear programming method and MATLAB® were used tosolve the WRDDM model, and the results were compared bothamong the years and the different regions.

4.1. Efficiency difference among years

4.1.1. Coal-fired power plantsTable 3 shows the notations used in this model. As mentioned

above, the CO2 emissions were derived from the energy con-sumption and caused the same value of b3 and b5.

The value of b for each province is displayed in Fig. 2. The in-efficiency among the provinces varies significantly. In addition, theaverage b each year was 0.473, 0.392, and 0.272, which implies thatthe differences among the provinces was narrowing over time. Forprovinces such as Liaoning, Jilin, Heilongjiang, Henan, Guangxi,Sichuan, Yunnan, Shanxi and Xinjiang, the b was higher than theaverage all three years. In contrast, provinces such as Hebei,Shanghai and Zhejiang outperformed the average in three years.For Hainan and Jiangxi, the sample size is considerably small (1 and2, respectively), and the b derived cannot be of great reference.

It should be noted that the comparison among years is notprecise since the samples for three years are not exactly the same.To make the b comparable in these years and to explore how effi-ciency has changed over time, we further screened a total of 261power plants with data for all the years as a new sample andmeasured the inefficiency of each input and output. The results areshown in Table 4.

Table 4 shows that the average b has dropped over the years.Specifically, approximately 62.45% of all the plants showed adescending trend in b, and 23.47% of them showed the oppositetrend. This indicates that most of the plants have improved theiroverall efficiency and that the gap among the plants was narrowing.It is interesting to find that in 2014, the b1 and b3 weremuch higherthan those in 2012 and 2013, which shows that the plants per-formed much worse in 2014 than in the last two years in terms oflabor and energy utilization. This could be explained by thedisparity of improvement and the diffusion of renewables in thepower generation portfolio among different plants. On the onehand, for some plants, the improvements in labor usage and energysavings were slower, which caused a larger gap between the plants.On the other hand, given that more renewable power was accessedto the grid, more thermal units needed to standby in case of thepower discontinuity. Therefore, the labor force may not be fullyutilized and the energy consumed may be wasted. These reasonshave caused the absolute value of b1 and b3 to become even higherthan before.

Moreover, b4 has significantly decreased, which implies that onaverage, the power generation capacity has achieved its optimum,

Table 1Data from coal-fired power enterprises.

Variables Unit Mean Std. Dev. Min Max

2012(n¼ 495)L (Labor) Person 711.49 789.88 12.00 10021.00K (Assets) Million yuan 4347.41 20810.37 14.34 460527.76E (energy consumption) Ton 1539777.93 1198101.32 67100.63 7536937.32Y (generation capacity) MWh 5120385.60 3990635.00 145602.70 25926000.00B (CO2) Ton 4237698.75 3317864.95 167744.52 20801947.00



Table 2Data from gas-fired power enterprises.

Variables Unit Mean Std. Dev. Min Max


2013(n¼ 35)L (Labor) Person 208.31 230.00 41 1199K (Assets) Million yuan 1678.65 999.81 294.92 4323.15E (energy consumption) Ton 550341.98 495712.94 34996.66 2506415.70Y (generation capacity) MWh 2334199.70 1607860.90 140000.00 6234800.00B (CO2) Ton 1397302.36 1364912.55 96590.78 6917707.33


0

5

10

15

20

25

30

35

40

45

50

Beijing

Tia njin

Hebe

iS hanxi

Inne

r…Lia

oning

Jilin

Heilon gj…

Shanghai

Ji angsu

Zhe jiang

Anhu

iFujian

Jiangxi

Shando

ngHe

nan

Hube

iHu

nan

Gua ngd…

G uangxi

Hainan

Chon

gqi…

Sichuan

Guizh

ouYun n

anS haanxi

Gansu

Qinghai

Ningxia

Xinjiang

numbe

rsof

coal-firedpo

wer

plan

ts

2012

2013

2014

Fig. 1. Provincial distribution of coal-fired power plants.


Table 3Notations and meanings.

Notations Meaning

b Overall inefficiency(improvement potential)b1 Labor inefficiencyb2 Asset-used inefficiencyb3 Energy-saving potentialb4 Output improvement potentialb5 Emission Abatement potential

0.0

0.1

0.2

0.3

0.4

0.5

Overallin

efficie

ncy

2012 2013 2014

Fig. 3. Value of the overall inefficiency, b, of each gas power plant.


and there is little room for further improvement. Thus, for thoseplants that have low values for b4, reducing energy consumptionand reducing the investment in labor and equipment are the pri-mary considerations if they want to promote overall efficiency.

4.1.2. Gas-fired power plantsThe sample of gas power plants was small and mainly located in

Beijing, Jiangsu, Zhejiang and Guangdong. Fig. 3 represents thevalue of b in ascending order.

Comparedwith the others, it can be seen that approximately 1/3to 1/2 of the power plants each year were globally efficient, whichmade b ¼ 0. For the other plants, the highest b each year decreasedover time. This indicates that the efficiency gap between the powerplants was narrowing and the plant owners had put more effortstoward upgrading their technologies or optimizing their inpututilization effectively.

Figs. 4e6 depict the energy consumption and b3 of each gas-fired plant from 2012 to 2014. It should be noted that all theplants in the figurewere independent of each other. The left verticalaxis represents energy consumption (coal-equivalence), and theright axis represents the energy saving potential. Similar to b, b3decreased, and nearly half of the plants were globally efficient. In2012, the highest b3 reached 0.7279; that is, compared with themost effective plants, for this plant, 72.79% of the energy

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

Beijing

Tia njin

Hebe

iShanxi

Inne

rMon

golia

Liaon

ing

Ji lin

Heilongji ang

Sha nghai

Jiangsu

Zhejiang

Anh u

iFujian

Jiangxi

Overallin

efficie

ncy

2012 201

Fig. 2. The average value of overall ine

Table 42012e2014 inefficiency of all variables.

Mean (Std. dev) b b1

2012 0.418(0.230) 0.264(0.230)2013 0.303(0.185) 0.189(0.229)2014 0.271(0.154) 0.431(0.295)

consumption could have been saved. The b3 values for 2013 and2014 were 0.6306 and 0.3336, respectively. Most of the plants hadtaken active measures to enhance the efficiency of power genera-tion according to the results.

4.2. Efficiency difference among regions

To examine the regional efficiency differences, we divided allthe provinces into six regions following the rule of CEC. Table 5 liststhe provinces included in each region, and Table 6 shows the meanand standard deviation of b and the rank of each province. The rankis according to the results of the overall efficiency, and the regionswith a higher overall efficiency will have higher ranks.

The overall efficiency has improved over time, which is consis-tent with previous results. Specifically, East and North China hadthe best performance in three years, which is the same as what Shiet al. (2010) discovered with data from 2000 to 2006. For EastChina, high efficiency may have resulted from the prosperouseconomy. With the highest GDP and most-developed industry andcommerce, East China has taken a leading role in China's

Shando

ngH e

nan

Hub e

iHu

n an

Guang don

gGu

angxi

Haina n

Cho n

g qi ng

Sichuan

G uizh

ouYunn

anShaanxi

Gansu

Qinghai

Ningxia

Xinjiang

Average

3 2014

fficiency, b, in different provinces.

b2 b3 b4

0.230(0.260) 0.232(0.265) 0.540(0.659)0.058(0.119) 0.335(0.281) 0.373(0.576)0.108(0.180) 0.436(0.293) 0.051(0.206)

0%

10%

20%

30%

40%

50%

60%

70%

80%

0

200

400

600

800

1,000

1,200

1,400

1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31Ene

rgy

cons

umpt

ion(

Tho

usan

dto

nsof

CE

/y)

op mal consump on excessive consump on Energy-saving poten al

Fig. 4. The energy consumption and energy-saving potential, b3, of each gas power plant in 2012.

0%

10%

20%

30%

40%

50%

60%

70%

80%

0

500

1,000

1,500

2,000

2,500

3,000

1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35Ene

rgy

cons

umpt

ion(

Tho

usan

dto

nsof

CE

/y)




development, and this has promoted investments in advancedtechnologies. For North China, which includes two municipalities,Beijing and Tianjin, policy-making and asset management werequite stringent, leading to better utilization in labor and capital.Additionally, the plants in Hebei stayed operational day and nightand contributed to higher efficiency without the energy lost in therestarting or closing process. The efficiencies in the Southwest andNortheast were the two lowest among the six regions. In theSouthwest, hydropower was the main source of energy, and ther-mal power mainly played the role of peak-shaving, especially inYunnan and Guizhou. In the Northeast, there were many smallthermal power plants that performed comparatively poorly due tothe lack of advanced management and technology. The total gen-eration hours were much lower than the national average (CECreport, 2016).

Furthermore, we sorted and located the plants whose b equaled0. Figs. 7e9 show the distribution of the globally efficient plantseach year. More than half of the plants were located in North andEast China, especially in Zhejiang Province. The remaining plants

were dispersedly in other regions. This is consistent with what wehave discussed above.

4.3. Factor analysis on efficiency

To examine if the size of the plants and the type of ownershiphave a significant impact on overall efficiency, we classified thesamples by different scales and ownership.

4.3.1. Scale differenceTotal installed capacity is a popular index for distinguishing

large and small plants. In this paper, we followed the rule by CECand combined the plants into two groups. Plants with less than1000MW installed capacity were defined as small, and those withmore than 1000MW installed capacity were defined as large. A boxplot was used to illustrate the differences between scales.

As shown in Fig. 10, the small plants had higher median valuesand variances than their large counterparts. Based on this finding,we assumed that there was a strong correlation between scale and

0%

10%

20%

30%

40%

50%

60%

70%

80%

0

200

400

600

800

1,000

1,200

1,400

1,600

1,800

1 3 5 7 9 11 13 15 17 19 21 23 25 27 29

Ene

rgy

c ons

umpt

ion(

Tho

usan

dto

nsof

CE

/y)



Table 5Classification of regions.

Regions Provinces included

North China Beijing, Tianjin,Hebei, Shanxi, Inner MongoliaNortheast Liaoning, Jilin,HeilongjiangEast Shanghai, Jiangsu,Zhejiang, Fujian,Anhui, Jiangxi,ShandongMiddle&South Henan, Hubei,Hunan, Guangxi,GuangdongSouthwest Chongqing, Sichuan,Guizhou, YunnanNorthwest Shaanxi, Gansu,Ningxia, Qinghai,Xinjiang


efficiency. Then, Kolmogorov-Smirnov (K-S) and Mann-Whitney(M-W) tests were used to confirm our assumption. Table 7 showsthe results of the two tests. They all refused the H0 hypothesis thatthe large plants and small plants share the same pattern on overallefficiency. Generally, the large plants performed better than thesmall plants. The large plants could benefit from various returns ofscale and well-developed management systems, which couldfurther improve the efficiency of labor and capital use.

4.3.2. Ownership difference

4.3.2.1. Different groups. The five groups (Huaneng, Datang, Hua-dian, Guodian and Zhongdiantou, or FGs) are the dominantelectricity-generating groups in China, and their annual generatingcapacity could account for approximately 50% of the total. Ac-cording to CEC industrial statistics, in 2013, the total generatingcapacity of the FGs was 441,004MW, comprising 51.1% of the totalgeneration. In 2014, the values were 433,967MW and 47.4%,respectively. Table 8 shows the number of different groups in the

Table 6b-value and ranks in different regions from 2012 to 2014.

2012 Rank 2013

North 0.4302(0.2461) 2 0.374Northeast 0.5485(0.2882) 6 0.493East 0.3521(0.2759) 1 0.319Middle& South 0.4309(0.2436) 3 0.400Southwest 0.4871(0.2570) 5 0.445Northwest 0.4778(0.2302) 4 0.421

sample.Before any testing, we assumed that the efficiency of the FGs

should be higher than the non-FGs for their sufficient funds,advanced technology, and well-developed management, but theresult shown in Fig. 11 negated our assumption. Similarly, we per-formed K-S and M-W tests, as shown in Table 9. Even under aconfidence level of 10%, the H0 hypothesis still cannot be rejected;that is, there was no significant difference in overall efficiency be-tween the FGs and the non-FGs.

To further determine the reasons, we compared the inefficiencyof each input and output. Fig. 12 shows the distribution of b1.

In 2012, there was not much efficiency difference between thetwo groups, but in 2013 and 2014, the b1 of the FGs were muchhigher than those of the non-FGs. This indicates that the labor forcein the non-FGs was utilized more efficiently than the labor force inthe FGs.

Table 10 shows the mean and standard deviation of b2. It can beseen that the FGs performed better in capital utilization.

The difference in b3 is reported in Fig. 13. In 2012, 2013, theenergy-saving potential in the FGs was higher, and in 2014, theyreversed this trend. This indicates that the FGs improved their ef-ficiency of energy consumption and exceeded the performance ofthe non-FGs.

Similarly, we selected the globally efficient plants and collectedthe groups towhich they belonged. Fig.14 shows the constitution ofthe different groups. In 2012, 2014, the FGs had more efficientplants, and in 2013, the opposite was true. It is difficult to distin-guish which group was better because of the fluctuation.

Rank 2014 Rank

2(0.2115) 2 0.2476(0.1250) 27(0.1750) 6 0.3294(0.1847) 57(0.2202) 1 0.2268(0.1345) 10(0.2606) 3 0.2903(0.1687) 40(0.2285) 5 0.3492(0.1456) 65(0.1906) 4 0.2620(0.1492) 3

Fig. 7. Distribution of the globally efficient plants (b¼0) in 2012.

Fig. 8. Distribution of the globally efficient plants (b¼0) in 2013.

Fig. 9. Distribution of globally efficient plants (b¼0) in 2014.

Fig. 10. The value distribution of b in two different scales from 2012 to 2014.

Table 7Results of the nonparametric tests.

K-S Test M-W Test

2012 0.500 (0.000) �10.547 (0.000)2013 0.604 (0.000) �12.558(0.000)2014 0.361 (0.000) �6.294(0.000)

Table 8Distribution of the FGs and non-FGs.

Five groups Non-five groups Proportion of Five groups

2012 290 205 58.59%2013 309 232 57.12%2014 198 124 61.49%

Fig. 11. The value distribution of b in the different groups from 2012 to 2014.



K-S Test M-W Test

2012 1.330(0.058) �1.521(0.128)2013 0.927(0.356) �0.960(0.337)2014 0.918(0.368) �0.746(0.456)

Fig. 12. The value distribution of b1 in the different groups from 2012 to 2014.

Table 10Mean and std. dev of b2 in the different groups.

Five groups Non-five groups

2012 0.113(0.017) 0.083(0.016)2013 0.286(0.028) 0.374(0.029)2014 0.042(0.010) 0.084(0.016)

Fig. 13. The value distribution of b3 in the different groups from 2012 to 2014.

Fig. 14. The constitution of the different groups for the best performing (b¼0) plants.

Fig. 15. The value distribution of b in the SOEs and the non-SOEs from 2012 to 2014.


K-S Test M-W Test

2012 1.355(0.050) �2.248(0.025)2013 1.749(0.004) �2.736(0.006)2014 1.684(0.007) �2.910(0.004)


Based on the results, the plants in the FGs outperformed in assetutilization, while the non-FGs had dominance in labor force usageand energy savings.

4.3.2.2. State-owned and nonstate-owned enterprises. We thendiscussed the correlation between efficiency and the state-ownedenterprises (SOEs)/non-SOEs. Fig. 15 and Table 11 show the distri-bution of b and the results of the nonparametric tests.

Under the 5% confidence level, the H0 hypothesis was rejected,and the b of the SOEs was higher than that of the non-SOEs. Basedon this, we compared the inefficiency of each specific input andoutput as we did before. Figs. 16 and 17 show the results.

We sorted the plants with full efficiency and determined theirownership, as shown in Fig. 18. It is interesting to find that onaverage, the SOEs had worse performance than the non-SOEs (seeFig. 16), but more than 70% of the globally efficient plants also

Fig. 17. The value distribution of b3 in the SOEs and the non-SOEs from 2012 to 2014.

Fig. 18. The distribution of the SOEs/non-SOEs for the best performing (b¼0) plants.

Fig. 16. The value distribution of b1 in the SOEs and the non-SOEs from 2012 to 2014.


belonged to the SOEs. This implies the great diversity in the effi-ciency of the SOEs.

Based on this finding, the labor inefficiency of the SOEs isconsiderably higher than that of the non-SOEs, which implies thatthe SOEs have a severe workforce surplus. The energy saving po-tential, b3, shows a similar pattern as b1. For capital utilization,there is no significant difference between the two groups.

5. Conclusions

This paper has offered a general view of the overall efficiency ofthermal power plants and shed light on the factor-specific potentialfor improvement. We found that East China outperformed theother regions because of its thriving economy and advanced tech-nologies. For Southwest and Northeast China, the low efficiencyresulted from the dominance of hydropower and small, poorlymanaged plants, respectively. This could be a reference for relevantpolicymaking. The provinces with poor performance can learnadvanced technology and management systems from the better-performing provinces or they can cooperate if those plantsspecialize in different parts and learn the expertise of another. Thiscould help take full advantage of the driving force of efficient powerplants and reduce the excessive use of fossil energy.

In the efficiency comparison among the years, we found that in2014, the overall efficiency was higher than that in the last twoyears, but the efficiency of labor and energy consumption wereworse. Considering that the efficiency measured in this paper wasrelative to the other plants in the sample, the efficiency differenceamong the years could be explained by the different paces ofmanagement promotion or the popularization of new technologiesgenerated in the different plants.

We also obtained interesting findings when studying the de-terminants of overall efficiency for the individual plants. Generally,the large plants performed better than the small plants. In China,small thermal power plants have often been established to balancethe uneven distribution of thermal energy. In this case, the smallplants were expected to aid in recognizing which factor contributesmost to inefficiency and improve it accordingly with the referenceof this model. If they have high efficiency in labor and asset utili-zation but poor efficiency in energy consumption, efforts should bemade in technical transformation. If possible, shutting some of theinefficient plants down and establishing new plants with muchhigher installed capacity could also be an effective alternative interms of emission abatement.

The plants in five groups (FGs) were thought to bemore efficientthan their counterparts for their dominance in power generationand technology investment. However, according to our study, therewas notmuch difference between the FGs and the non-FGs in termsof overall efficiency. Although the plants in the FGs had high effi-ciency in asset utilization and invested more in technologies, theyhad poor performance in labor management, which weighed downthe overall efficiency. This pattern could also be applied to the SOEsand the non-SOEs.

There are still some drawbacks in this research. The inefficiencywe obtained from the DEA model is a relative concept, which ishighly influenced by the sample selection and extreme values,therefore, the results may be somehow affected, especially whensome efficient plants are missing. Furthermore, due to data limi-tations, we abandoned plants with incomplete data, which resultedin a small sample capacity and distortions. In this case, some slightdeviation may exist in our analysis.

Acknowledgement

The authors gratefully acknowledge financial support from the


National Natural Science Foundation of China, Nos. 71673019,71690245, 71210005, 71503242, 71273253, and the EnvironmentalDefense Fund. All remaining errors are the sole responsibility of theauthors.

References

Chen, P.C., Yu, M., Managi, S., Chang, C., 2011. Non-Radial Directional PerformanceMeasurement with Undesirable Outputs. Tohoku University Working Paper,Japan.

Chen, P.C., Yu, M., Chang, C., Hsu, S., Managi, S., 2015. The enhanced Russell-baseddirectional distance measure with undesirable outputs: numerical exampleconsidering CO2 emissions. Omega 53, 30e40.

Choi, Y., Zhang, N., Zhou, P., 2012. Efficiency and abatement costs of energy-relatedCO2 emissions in China: a slacks-based efficiency measure. Appl. Energy 98,198e208.

Chung, Y.H., F€are, R., Grosskopf, S., 1997. Productivity and undesirable outputs: adirectional distance function approach. J. Environ. Manag. 51, 229e240.

F€are, R., Grosskopf, S., Pasurka, C.A., 2007. Environmental production functions andenvironmental directional distance functions. Energy 32, 1055e1066.

Fukuyama, H., Weber, W.L., 2009. A directional slacks-based measure of technicalinefficiency. Soc. Econ. Plann. Sci. 43, 274e287.

Han, Y.M., Long, C., Geng, Z.Q., Zhang, K.Y., 2018. Carbon emission analysis andevaluation of industrial departments in China: an improved environmental DEAcross model based on information entropy. J. Environ. Manag. 298e307, 2015.

International Energy Agency (IEA), 2018. CO2 Emissions from Fuel Combustion2018. IEA/OECD, Paris.

Jerry, B., Eva, H., Robert, L., 2012. Energy efficiency and policy in Swedish pulp andpaper mills: a data envelopment analysis approach. Energy Policy 42, 569e579.

Li, H.Z., Zou, T., 2012. The efficiency and influencing factors of power generationindustry in China during 2000-2009. Revolution 10, 44e50.

Liu, Y.N., Wang, K., 2015. Energy efficiency of China's industry sector: an adjustednetwork DEA-based decomposition analysis. Energy 93, 1328e1337.

Mandal, S.K., 2010. Do undesirable output and environmental regulation matter inenergy efficiency analysis? Evidence from Indian cement industry. EnergyPolicy 38, 6076e6083.

María, M., Germ�an, G., Trinidad, G., Rafael, C., Ram�on, S., 2016. Eco-efficiencyassessment of wastewater treatment plants using a weighted Russell direc-tional distance model. J. Clean. Prod. 137, 1066e1075.

Niu, X.M., 2017. Research on Regional Difference and Convergence of China's Carbon

Emission Efficiency from the Perspective of Total Factors. Southwestern Uni-versity of Finance and Economics, China.

Shi, G.M., Bi, J., Wang, J.N., 2010. Chinese regional industrial energy efficiencyevaluation based on a DEA model of fixing non-energy inputs. Energy Policy 38,6172e6179.

Sueyoshi, T., Goto, M., 2011. DEA approach for unified efficiency measurement:assessment of Japanese fossil fuel power generation. Energy Econ. 33, 292e303.

Teng, F., Wu, Z.X., 2003. Performance Analysis on Chinese electric power enter-prises. J. Quant. Tech. Econ. 6, 127e130 (China).

Wang, Q.W., Hang, Y., Sun, L.C., Zhao, Z.Y., 2016. Two-stage innovation efficiency ofnew energy enterprises in China: a non-radial DEA approach. Technol. Forecast.Soc. Change 112, 254e261. https://doi.org/10.1016/j.techfore.2016.04.019.

Wei, C., Andreas, L., Liu, B., 2013. An empirical analysis of the CO2 shadow price inChinese thermal power enterprises. Energy Econ. 44, 22e31.

Wei, C., Andreas, L., Liu, B., 2015. Energy-saving and emission-abatement potentialof Chinese coal-fired power enterprise: a non-parametric analysis. Energy Econ.49, 33e43.

Xie, B.C., Fan, Y., Qu, Q.Q., 2012. Does generation form influence environmentalefficiency performance? An analysis of China's power system. Appl. Energy 96,261e271.

Yang, H.L., Michael, P., 2009. Incorporating both undesirable outputs and uncon-trollable variables into DEA: the performance of Chinese coal-fired powerplants. Eur. J. Oper. Res. 197, 1095e1105.

Yang, H.L., Michael, P., 2010. The necessity of distinguishing weak and strongdisposability among undesirable outputs in DEA: environmental performanceof Chinese coal-fired power plants. Energy Policy 38, 4440e4444.

Zhang, G.X., Xia, D.W., 2011. Ownership structure, environment regulation and ef-ficiency of Chinese power generation industryddstochastic frontier produc-tion function analysis based on provincial panel data during 2003d2009. ChinaInd. Econ. 6, 130e140.

Zhang, Ning, Choi, Yongrok, 2014. A note on the evolution of directional distancefunction and its development in energy and environmental studies 1997e2013.Ren. Sus. Energy Rev. 33, 50e59.

Zhang, Ning, Choi, Yongrok, 2013. A comparative study of dynamic changes in CO2emission performance of fossil fuel power plants in China and Korea. EnergyPol. 62, 324e332.

Zhou, P., Ang, B.W., Poh, K.L., 2008. A survey of data envelopment analysis in energyand environmental studies. Eur. J. Oper. Res. 189, 1e18.

Zhou, P., Ang, B.W., Wang, H., 2012. Energy and CO2 emission performance inelectricity generation: a non-radial directional distance function approach. Eur.J. Oper. Res. 221, 625e635.

http://refhub.elsevier.com/S0959-6526(19)30769-3/sref2




























































https://doi.org/10.1016/j.techfore.2016.04.019











































Documents

Journal of Cleaner Production - ae.ruc.edu.cn